Real-time monitoring and early warning system and method for data of lithium battery of electric bicycle

Abstract

The invention discloses an electric bicycle lithium battery data real-time monitoring and early warning system and method, and relates to the technical field of battery management, and the system comprises a multi-dimensional data collection module which is used for obtaining a multi-source heterogeneous data set of a lithium battery system; the collaborative feature extraction module is used for generating a comprehensive evaluation parameter set; the dynamic threshold generation module is used for constructing a self-adaptive early warning boundary model according to the comprehensive evaluation parameter set; the intelligent decision module is used for generating a hierarchical control instruction set based on a multi-objective optimization algorithm; and the cloud collaboration module is used for synchronizing the hierarchical control instruction set to the edge computing node and the cloud management platform, and triggering a multi-level linkage protection mechanism based on the game theory when the thermal runaway risk index is detected to exceed a first dynamic threshold value. According to the electric bicycle lithium battery data real-time monitoring and early warning system and method provided by the invention, the safety and reliability of a battery system are improved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of battery management, in particular to a real-time monitoring and early warning system and method for lithium battery data of an electric bicycle. BACKGROUND

[0002] At present, the lithium battery management system mainly relies on basic parameters such as voltage and current for state monitoring, and it is difficult to comprehensively reflect the real working condition of the battery. The traditional method adopts a fixed threshold and a single model, which leads to problems such as large error in health assessment, lag in thermal runaway early warning, etc. Especially in the case of battery aging or extreme environment, the accuracy of the existing technology decreases significantly.

[0003] In addition, the existing control strategy lacks dynamic adjustment capability and cannot adapt to the performance degradation of the battery and the change of the environment. The separate optimization of each subsystem leads to poor overall energy efficiency, and the localized architecture also limits the fault response speed. These problems seriously restrict the improvement of the safety and reliability of the battery management system. SUMMARY

[0004] In order to overcome the deficiencies of the prior art, the application provides a real-time monitoring and early warning system and method for lithium battery data of an electric bicycle.

[0005] In a first aspect, the application provides a real-time monitoring and early warning system for lithium battery data of an electric bicycle, which comprises: A multi-dimensional data acquisition module is configured to acquire a multi-source heterogeneous data set of a lithium battery system, wherein the multi-source heterogeneous data set comprises a single cell voltage ripple spectrum, a battery pack temperature field distribution matrix, a charge and discharge current harmonic component, and a battery internal resistance dynamic change curve. A collaborative feature extraction module is configured to perform cross-domain feature alignment on the multi-source heterogeneous data set based on a space-time fusion algorithm, and generate a comprehensive evaluation parameter set comprising a health state score, a thermal runaway risk index, and a residual life prediction value. A dynamic threshold generation module is configured to construct an adaptive early warning boundary model according to the comprehensive evaluation parameter set, and calculate a voltage equalization compensation coefficient, a maximum allowable charging rate correction factor, and a thermal gradient suppression parameter through a double-layer neural network. An intelligent decision-making module is configured to generate a hierarchical control instruction set based on a multi-objective optimization algorithm, wherein the instruction set comprises a pulse charging duty cycle adjustment amount, an active equalization topology switching strategy, a cooling fan speed curve, and a battery state of charge calibration parameter. A cloud collaboration module is configured to synchronize the hierarchical control instruction set to an edge computing node and a cloud management platform, and trigger a multi-level linkage protection mechanism based on game theory when the thermal runaway risk index exceeds a first dynamic threshold.

[0006] Preferably, the collaborative feature extraction module comprises: a ripple feature analysis unit configured to perform wavelet transform-Hilbert joint analysis on the monomer voltage ripple spectrum to extract high-frequency oscillation energy accumulation value and low-frequency drift component, and to distinguish normal charging and discharging ripple from abnormal oscillation caused by early failure by establishing a voltage fluctuation energy density spectrum, wherein a preliminary warning is triggered when the high-frequency oscillation energy accumulation value exceeds a preset reference value; a thermal field modeling unit configured to establish a three-dimensional heat conduction equation of the temperature field distribution matrix, solve the internal thermal flow distribution of the battery pack by using a finite element method, and calculate the critical heat flux density under the maximum temperature difference constraint, and to start an active cooling strategy when the critical heat flux density threshold of a local hot spot area exceeds a preset threshold by comparing the deviation degree of the actual heat flux density and the critical value in real time; an impedance spectrum analysis unit configured to fit the relaxation time constant of the internal resistance dynamic change curve by using a least squares method, and to solve by using a Levenberg-Marquardt optimization algorithm, and to construct a Cole-Cole spectrum including a plurality of characteristic frequency points, and to realize dynamic evaluation of the battery aging state by monitoring the change rate of the relaxation time constant, and to determine abnormal aging when the relaxation time constant mutation value exceeds twice the standard deviation of the historical mean value; a feature fusion subunit configured to use an attention mechanism to weight and fuse the above three types of features, and to dynamically adjust the weight coefficients according to the current working condition.

[0007] Preferably, the dynamic threshold generation module comprises: a nonlinear mapping unit configured to construct a two-dimensional decision plane including state of health score SOH and thermal runaway risk index R_th, and to generate a dynamic warning boundary surface; the surface divides the working state into three regions of a safe region, a warning region, and a dangerous region, and starts a derating operation mode when a data point enters the warning region; an adaptive learning unit configured to use a long short-term memory network (LSTM) to construct a double-layer neural network, and to update the weight matrix W of the double-layer neural network online according to the battery degradation trajectory, and to introduce a ripple energy dynamic adjustment factor into the update rule; the loss function comprises voltage consistency error, temperature gradient error, and capacity attenuation error; the adaptive learning unit performs global parameter update every preset number of charging and discharging cycles, and retains a historical optimal weight copy as a rollback backup at the same time; a threshold optimization subunit configured to generate group random working condition data by using Monte Carlo simulation, and to optimize each warning threshold combination by using a particle swarm algorithm.

[0008] Preferably, the intelligent decision-making module comprises: A multi-objective optimization unit is configured to construct a dynamic optimization space in the dimensions of voltage balance, temperature stability, and battery life loss, automatically adjust optimization weights according to a user-set performance priority, safety priority, or life priority mode, and output an optimal control parameter combination. A pulse charging control unit is configured to use an adaptive fuzzy proportional-integral-derivative (PID) algorithm to adjust a charging current waveform in real time, dynamically select a plurality of preset pulse modes according to a battery temperature and a state of health, and include an intelligently adjusted charging period and a relaxation interval in each pulse period. An equalization strategy decision unit is configured to intelligently switch between active equalization and passive equalization modes by monitoring a battery pack voltage difference and a temperature gradient in real time, wherein the active equalization uses a distributed energy transfer architecture based on a switched capacitor matrix, and an equalization accuracy is controlled within a preset range. A thermal management control unit is configured to generate a hierarchical heat dissipation strategy according to a temperature field analysis result, control a speed of a heat dissipation fan through pulse width modulation (PWM) speed control, and start a directional strong cooling mode when a local hot spot is detected.

[0009] Preferably, the cloud collaboration module includes: An edge-cloud data synchronization unit is configured to use a differential compression technique to achieve bidirectional real-time synchronization of control instructions and state data, and ensure instruction consistency between an edge node and a cloud platform. A multi-level linkage arbitration unit is configured to, when a thermal runaway risk exceeds a limit, construct a three-party collaborative decision-making model of a battery pack, a charging pile, and a vehicle-mounted system based on game theory, and dynamically allocate protection response priorities of the systems. A digital twin mirror unit is configured to construct a virtual model that is completely synchronized with a physical battery in the cloud, and use the virtual model to test a safety margin of different control strategies in large-scale parallel simulation. An emergency broadcast unit is configured to send a risk warning to associated devices and trigger a collaborative protection mechanism of surrounding systems.

[0010] In a second aspect, a real-time monitoring and early warning method for a lithium battery of an electric bicycle includes: A multi-source heterogeneous data set of a lithium battery system is obtained, and the multi-source heterogeneous data set includes a single-cell voltage ripple spectrum, a battery pack temperature field distribution matrix, a charging and discharging current harmonic component, and a battery internal resistance dynamic change curve. A cross-domain feature alignment is performed on the multi-source heterogeneous data set based on a space-time fusion algorithm, and a comprehensive evaluation parameter set including a state of health score, a thermal runaway risk index, and a remaining life prediction value is generated. An adaptive early warning boundary model is constructed according to the comprehensive evaluation parameter set, and a voltage balance compensation coefficient, a maximum allowable charging rate correction factor, and a thermal gradient suppression parameter are calculated through a double-layer neural network. The hierarchical control instruction set is generated based on a multi-objective optimization algorithm, and the instruction set includes a pulse charging duty cycle adjustment amount, an active balancing topology switching strategy, a cooling fan rotating speed curve, and a battery state of charge calibration parameter. The hierarchical control instruction set is synchronized to an edge computing node and a cloud management platform, and when a thermal runaway risk index is detected to exceed a first dynamic threshold, a multi-level linkage protection mechanism based on game theory is triggered.

[0011] Preferably, the multi-source heterogeneous data set is aligned across domains based on a space-time fusion algorithm to generate a comprehensive evaluation parameter set including a health state score, a thermal runaway risk index, and a residual life prediction value, specifically: The monomer voltage ripple spectrum is subjected to wavelet transform-Hilbert joint analysis to extract a high-frequency oscillation energy accumulation value and a low-frequency drift component, and the ripple feature analysis unit distinguishes normal charging and discharging ripples from abnormal oscillations caused by early faults by establishing a voltage fluctuation energy density spectrum, wherein a preliminary warning is triggered when the high-frequency oscillation energy accumulation value exceeds a preset reference value. A three-dimensional heat conduction equation of the temperature field distribution matrix is established, a finite element method is used to solve the internal heat flow distribution of the battery pack, and the critical heat flux density under the maximum temperature difference constraint is calculated, and the thermal field modeling unit starts an active cooling strategy when the critical heat flux density threshold of the local hot spot area exceeds the preset threshold by real-time comparison of the deviation degree of the actual heat flux density and the critical value; The relaxation time constant of the internal resistance dynamic change curve is fitted by the least squares method, and the Levenberg-Marquardt optimization algorithm is used for solving, and the impedance spectrum analysis unit synchronously constructs a Cole-Cole spectrum including multiple characteristic frequency points, and realizes dynamic evaluation of the battery aging state by monitoring the change rate of the relaxation time constant, and determines that the battery is abnormally aged when the relaxation time constant mutation value exceeds twice the standard deviation of the historical mean value. The above three types of features are weighted and fused using an attention mechanism, and the weight coefficients are dynamically adjusted according to the current working condition.

[0012] Preferably, an adaptive early warning boundary model is constructed according to the comprehensive evaluation parameter set, and a double-layer neural network is used to calculate a voltage equalization compensation coefficient, a maximum allowed charging rate correction factor, and a thermal gradient suppression parameter, specifically: A two-dimensional decision plane including a health state score SOH and a thermal runaway risk index R_th is constructed to generate a dynamic early warning boundary surface; the surface divides the working state into three regions: a safe region, a warning region, and a dangerous region, and when a data point enters the warning region, a derating operation mode is started. A long short-term memory (LSTM) network is used to construct a double-layer neural network, a weight matrix W of which is updated online according to a battery degradation trajectory, and an update rule of which introduces a ripple energy dynamic adjustment factor; the loss function includes voltage consistency error, temperature gradient error and capacity attenuation error; the adaptive learning unit performs global parameter updating every preset charging and discharging cycle, while retaining a historical optimal weight copy as a rollback backup; Random working condition data of the group is generated by Monte Carlo simulation, and each early warning threshold combination is optimized by using a particle swarm algorithm.

[0013] Preferably, a hierarchical control instruction set is generated based on a multi-objective optimization algorithm, and the instruction set includes a pulse charging duty cycle adjustment amount, an active balancing topology switching strategy, a cooling fan speed curve and a battery state of charge calibration parameter, and specifically comprises the following steps: A dynamic optimization space is constructed in the dimensions of voltage uniformity, temperature stability and battery life loss, and the optimization weight is automatically adjusted according to the user-set performance priority, safety priority or life priority mode to output an optimal control parameter combination; An adaptive fuzzy proportional-integral-derivative (PID) algorithm is used to adjust the charging current waveform in real time, and a plurality of preset pulse modes are dynamically selected according to the battery temperature and health state, and each pulse period includes an intelligently adjusted charging period and a relaxation interval; The battery pack voltage difference and temperature gradient are monitored in real time, and intelligent switching is performed between active balancing and passive balancing modes, wherein the active balancing adopts a distributed energy transfer architecture based on a switched capacitor matrix, and the balancing accuracy is controlled within a preset range; A hierarchical cooling strategy is generated according to the temperature field analysis result, the speed of the cooling fan is controlled by pulse width modulation (PWM), and a directional strong cooling mode is started when a local hot spot is detected.

[0014] Preferably, the steps of generating a hierarchical cooling strategy according to the temperature field analysis result, controlling the speed of the cooling fan by pulse width modulation (PWM), and starting a directional strong cooling mode when a local hot spot is detected, specifically comprise the following steps: Differential compression technology is used to realize bidirectional real-time synchronization of control instructions and state data, and ensure the consistency of instructions of edge nodes and the cloud platform; When the thermal runaway risk exceeds the limit, a three-party collaborative decision-making model of the battery pack, the charging pile and the vehicle-mounted system is constructed based on game theory, and the protection response priority of each system is dynamically allocated; A virtual model completely synchronized with the physical battery is constructed in the cloud, which is used for large-scale parallel simulation to test the safety margin of different control strategies; A risk warning is sent to the associated equipment to trigger the collaborative protection mechanism of the surrounding system.

[0015] Compared with the prior art, the present application has the following characteristics and beneficial effects: The multi-dimensional data acquisition module obtains single cell voltage ripple spectrum, battery pack temperature field distribution matrix, charge and discharge current harmonic components and internal resistance dynamic change curve of the lithium battery system, breaks through the limitation of traditional battery management system relying on a single data source, realizes comprehensive and accurate perception of battery electrochemical characteristics, thermodynamic state and electrical parameters, and provides high timeliness and high precision data basis for subsequent analysis; The spatio-temporal fusion algorithm based on the collaborative feature extraction module performs cross-domain feature alignment and deep correlation analysis on the multi-source heterogeneous data, generates a comprehensive evaluation parameter set containing health state score, thermal runaway risk index and residual life prediction value, overcomes the inherent defects of traditional static threshold and single model, and significantly improves the accuracy and environmental adaptability of battery state evaluation; The adaptive warning boundary model constructed by the dynamic threshold generation module combines with the double-layer neural network to dynamically calculate the voltage equalization compensation coefficient, the maximum allowed charge rate correction factor and the thermal gradient suppression parameter, effectively solves the poor adaptability problem of fixed threshold strategy under battery aging and extreme working conditions, and realizes intelligent optimization of the safety boundary; With the multi-objective optimization algorithm of the intelligent decision module, a hierarchical control instruction set containing pulse charge duty cycle adjustment amount, active equalization topology switching strategy, cooling fan speed curve and SOC calibration parameter is generated, which optimizes the energy efficiency ratio and life decay rate while ensuring system safety; Finally, the cloud-end edge collaborative execution of the control instruction is realized through the cloud collaborative module, and the multi-level linkage protection mechanism is triggered based on game theory, which improves the safety and reliability of the battery system, and provides an expandable intelligent solution for large-scale energy storage applications, which improves the battery management accuracy, system safety and comprehensive energy efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 is a structural block diagram of a lithium battery data real-time monitoring and early warning system for an electric bicycle mainly embodied by the present embodiment.

[0017] Figure 2 is a step block diagram of a lithium battery data real-time monitoring and early warning method for an electric bicycle mainly embodied by the present embodiment. DETAILED DESCRIPTION

[0018] The present application will be further described in detail below in conjunction with the following embodiments.

[0019] Referring to Figure 1 , the lithium battery data real-time monitoring and early warning system for an electric bicycle includes the following modules: A multi-dimensional data acquisition module is configured to acquire a multi-source heterogeneous data set of the lithium battery system, and the multi-source heterogeneous data set includes a single cell voltage ripple spectrum, a battery pack temperature field distribution matrix, a charge and discharge current harmonic component, and a battery internal resistance dynamic change curve. A collaborative feature extraction module is configured to perform cross-domain feature alignment on the multi-source heterogeneous data set based on a space-time fusion algorithm, and generate a comprehensive evaluation parameter set including a health state score, a thermal runaway risk index, and a residual life prediction value. A dynamic threshold generation module is configured to construct an adaptive early warning boundary model according to the comprehensive evaluation parameter set, and calculate a voltage balance compensation coefficient, a maximum allowable charge rate correction factor, and a thermal gradient suppression parameter through a double-layer neural network. An intelligent decision-making module is configured to generate a hierarchical control instruction set based on a multi-objective optimization algorithm, and the instruction set includes a pulse charge duty cycle adjustment amount, an active balancing topology switching strategy, a cooling fan speed curve, and a battery state of charge calibration parameter. A cloud collaboration module is configured to synchronize the hierarchical control instruction set to an edge computing node and a cloud management platform, and trigger a multi-level linkage protection mechanism based on game theory when the thermal runaway risk index exceeds a first dynamic threshold.

[0020] Firstly, the multi-dimensional data acquisition module obtains the single cell voltage ripple spectrum, battery pack temperature field distribution matrix, charge and discharge current harmonic component and internal resistance dynamic change curve and other multi-source heterogeneous data of the lithium battery system, breaking through the limitation of traditional battery management system relying on single data source, realizing comprehensive and accurate perception of battery electrochemical characteristics, thermodynamic state and electrical parameters, providing high timeliness and high precision data basis for subsequent analysis; the spatio-temporal fusion algorithm based on the collaborative feature extraction module performs cross-domain feature alignment and deep correlation analysis on multi-source heterogeneous data, generates a comprehensive evaluation parameter set containing health state score, thermal runaway risk index and residual life prediction value, overcomes the inherent defects of traditional static threshold and single model, significantly improves the accuracy and environmental adaptability of battery state evaluation; the adaptive warning boundary model constructed by the dynamic threshold generation module combines the double-layer neural network to dynamically calculate the voltage equalization compensation coefficient, the maximum allowed charge rate correction factor and the thermal gradient suppression parameter, effectively solves the poor adaptability problem of fixed threshold strategy under battery aging and extreme working conditions, realizes intelligent optimization of safety boundary; with the help of the multi-objective optimization algorithm of the intelligent decision module, the hierarchical control instruction set containing pulse charge duty cycle adjustment amount, active equalization topology switching strategy, cooling fan speed curve and SOC calibration parameters is generated, which optimizes the energy efficiency ratio and life decay rate while ensuring system safety; finally, the cloud-end collaborative module realizes the cloud-end edge collaborative execution of control instructions, and based on game theory triggers multi-level linkage protection mechanism, improves the safety and reliability of the battery system, and provides an expandable intelligent solution for large-scale energy storage applications, which improves the battery management precision, system safety and comprehensive energy efficiency.

[0021] Specifically, the working process of the multi-dimensional data acquisition module can include: real-time acquisition of transient fluctuation signals of battery cell voltage through a distributed sensor network, conversion of analog signals to digital waveform data using a high-precision ADC converter, and elimination of environmental electromagnetic interference using digital filtering technology. For temperature field monitoring, a three-dimensional matrix arrangement of temperature sensor arrays is used to capture temperature gradient changes at the surface and internal key nodes of the battery pack with millisecond-level sampling frequency. The charge and discharge current harmonic component is collected by a wideband Hall sensor to obtain the original current waveform, and the fundamental wave and harmonic components are separated by a fast Fourier transform (FFT) algorithm to accurately quantify the current distortion. The battery internal resistance detection uses the alternating current injection method, which applies a small-amplitude high-frequency excitation signal to the battery and synchronously measures the voltage response phase difference, and combines impedance spectrum analysis technology to invert the internal electrochemical impedance characteristics of the battery. These technical means cooperatively realize the stereoscopic perception of the battery electro-thermal-chemical multi-physical field parameters, providing full-dimensional data support for subsequent analysis.

[0022] The collaborative feature extraction module can include the following sub-units: The ripple characteristic analysis unit is used to perform a wavelet transform-Hilbert joint analysis on the monomer voltage ripple spectrum according to the following formula to extract the high-frequency oscillation energy accumulation value and low-frequency drift component: in, is the Hilbert-wavelet transform operator, is the instantaneous voltage of the kth battery, is the sampling time interval, N is the total number of battery cells, The ripple characteristic analysis unit distinguishes normal charge and discharge ripples from abnormal oscillations caused by early faults by establishing a voltage fluctuation energy density spectrum. When the high-frequency oscillation energy accumulation value exceeds a preset reference value, a primary warning is triggered. The thermal field modeling unit is used to establish the three-dimensional heat conduction equation of the temperature field distribution matrix, use the finite element method to solve the heat flow distribution inside the battery pack, and calculate the critical heat flux density under the maximum temperature difference constraint: in, is the critical heat flux threshold, is the anisotropic thermal conductivity tensor, is the density of the composite material, is the temperature-dependent specific heat capacity; The temperature value at the three-dimensional space coordinate (x, y, z) is calculated. The thermal field modeling unit compares the deviation between the actual heat flux density and the critical value in real time, and activates the active heat dissipation strategy when the critical heat flux density threshold of the local hot spot exceeds the preset threshold. The impedance spectrum analysis unit is used to fit the relaxation time constant of the dynamic change curve of the internal resistance by the least squares method, and the Levenberg-Marquardt optimization algorithm is used to solve: in, is the relaxation time constant, for Always measure the internal resistance value. is the initial internal resistance reference value, M is the total number of sampling points, To minimize the parameter values ​​of the objective function, the impedance spectrum analysis unit simultaneously constructs a Cole-Cole spectrum including multiple characteristic frequency points. By monitoring the change rate of the relaxation time constant, a dynamic assessment of the battery aging state is achieved. When the sudden change in the relaxation time constant exceeds two times the standard deviation of the historical mean, it is judged as abnormal aging. The feature fusion subunit is used to perform weighted fusion of the above three types of features using the attention mechanism, and the weight coefficient is dynamically adjusted according to the current working conditions.

[0023] Specifically, the ripple feature analysis unit performs multi-scale decomposition on the voltage fluctuation signal through wavelet transform, extracts the energy distribution characteristics of different frequency bands, and then constructs a time-frequency joint distribution map combined with Hilbert transform, effectively distinguishing the charge-discharge ripple under normal working conditions from the fault oscillation mode under abnormal working conditions. The thermal field modeling unit is based on the three-dimensional unsteady heat conduction equation, considering the anisotropic thermal conductivity characteristics of the battery pack composite material, using the finite element iterative algorithm to solve the temperature field distribution, and identifying the potential thermal runaway risk area through the heat flux vector field analysis. The impedance spectrum analysis unit uses a nonlinear least squares fitting algorithm combined with the Levenberg-Marquardt optimization method to extract the relaxation time constant representing the battery aging state from the internal resistance dynamic change curve, and simultaneously constructs the Cole-Cole spectrum to analyze the dynamic characteristics of the electrochemical interface. The feature fusion sub-unit introduces an attention mechanism to dynamically adjust the weight coefficients of each feature parameter according to the current working condition of the battery, such as increasing the decision weight of the temperature feature during high-rate charging and strengthening the contribution of the impedance spectrum feature during the battery aging stage, thereby realizing adaptive deep correlation of multi-source heterogeneous data.

[0024] The specific dynamic threshold generation module can include the following sub-units: A nonlinear mapping unit is used to construct a two-dimensional decision plane including the state of health score SOH and the thermal runaway risk index R_th, and to generate a dynamic warning boundary surface by solving the following partial differential equation: wherein, SOH is the current battery state of health score, SOH0 is the initial state of health reference value, R_th is the thermal runaway risk index, R_th0 is the risk reference value, k is the material aging sensitivity coefficient, k is the thermal stability coefficient, k is the temperature influence factor, T_max is the current maximum temperature, T_ref is the reference temperature; the surface divides the working state into three regions: safe zone, warning zone and danger zone, and starts the derating mode when the data point enters the warning zone; An adaptive learning unit is used to construct a double-layer neural network using a long short-term memory network (LSTM), and the weight matrix W is updated online according to the battery degradation trajectory, and the update rule introduces a ripple energy dynamic adjustment factor: wherein, W_new is the updated weight matrix, W_old is the weight matrix before updating, η is the dynamic learning rate, is the gradient of the loss function with respect to the weight, a hyperbolic tangent activation function, a ripple energy reference value, L is a loss function; the loss function includes voltage consistency error, temperature gradient error and capacity attenuation error; the adaptive learning unit performs global parameter update every preset number of charge and discharge cycles, while retaining a copy of the historical optimal weight as a rollback backup; a threshold optimization subunit for generating group random working condition data through Monte Carlo simulation, and optimizing each early warning threshold combination using a particle swarm algorithm, with a target function being: wherein, is a first weight, is a second weight, is a third weight, is a false alarm rate, is a missed alarm rate, is a response time index.

[0025] Specifically, the nonlinear mapping unit solves the partial differential equation coupling the battery state of health and thermal runaway risk to construct a dynamic early warning boundary surface with curvature characteristics in a two-dimensional decision plane. This surface takes into account the material property degradation caused by battery aging (such as the decrease in lithium ion diffusion coefficient due to SEI film thickening) and the influence of environmental temperature changes on thermal stability, so that the division of the safety boundary can adapt to the actual state of the battery in real time. The adaptive learning unit uses a double-layer LSTM network architecture, and the hidden layer neuron weight matrix is updated online by introducing a ripple energy dynamic adjustment factor. This mechanism enables the neural network to adapt to the progressive degradation of battery performance while retaining the historical optimal weight as a benchmark for fault recovery. The threshold optimization subunit generates a virtual dataset covering different aging stages and extreme working conditions through Monte Carlo simulation, and uses an improved particle swarm algorithm for multi-objective optimization in the false alarm rate, missed alarm rate and response time dimensions, finally obtaining a dynamic threshold combination that balances safety and economy.

[0026] The specific intelligent decision-making module can include the following subunits: a multi-objective optimization unit for constructing a dynamic optimization space in the dimensions of voltage uniformity, temperature stability and battery life loss, automatically adjusting the optimization weights according to the user's performance priority, safety priority or life priority mode, and outputting the optimal control parameter combination; a pulse charging control unit for using an adaptive fuzzy proportional-integral-derivative (PID) algorithm to adjust the charging current waveform in real time, dynamically selecting multiple preset pulse modes according to the battery temperature and state of health, and each pulse period including an intelligently adjusted charging period and a relaxation interval; An equalization strategy decision unit is configured to intelligently switch between active equalization and passive equalization modes by monitoring the battery pack voltage difference and temperature gradient in real time, wherein the active equalization adopts a distributed energy transfer architecture based on a switch capacitor matrix, and the equalization accuracy is controlled within a preset range. A thermal management control unit is configured to generate a hierarchical heat dissipation strategy according to the temperature field analysis result, control the rotation speed of a heat dissipation fan through pulse width modulation (PWM) speed control, and start a directional strong cooling mode when a local hot spot is detected.

[0027] Specifically, the multi-objective optimization unit constructs a multi-dimensional optimization space containing voltage equalization degree, temperature uniformity and capacity attenuation rate, adopts a Pareto frontier search algorithm to find an optimal solution set, and automatically adjusts the optimization target weight according to the user's set priority mode (such as safety priority or life priority). The pulse charging control unit develops an adaptive fuzzy PID controller, dynamically selects the amplitude, frequency and relaxation time combination mode of the charging pulse by real-time analysis of the battery temperature, state of health and current state of charge, suppresses the growth of lithium dendrites while improving the charging efficiency. The equalization strategy decision unit designs a hybrid equalization architecture, adopts a lower energy consumption passive equalization mode when the voltage difference is small, automatically switches to an active equalization mode based on a switch capacitor matrix when significant inconsistency is detected, and realizes optimal energy transfer by intelligently adjusting the equalization current. The thermal management control unit establishes a transfer function model of temperature-heat dissipation efficiency, generates a PWM speed curve of the heat dissipation fan using a model predictive control (MPC) algorithm, and starts a directional strong cooling mode when a local hot spot is identified, and adjusts the angle of the air duct guide plate to achieve accurate heat dissipation.

[0028] The specific cloud collaboration module can include the following sub-units: An edge-cloud data synchronization unit is configured to realize bidirectional real-time synchronization of control instructions and state data using differential compression technology to ensure the consistency of instructions between the edge node and the cloud platform. A multi-level linkage arbitration unit is configured to, when the thermal runaway risk exceeds the limit, construct a three-party collaborative decision-making model based on game theory among the battery pack, the charging pile and the vehicle-mounted system, and dynamically allocate the protection response priority of each system. A digital twin mirror unit is configured to construct a virtual model completely synchronized with the physical battery in the cloud for large-scale parallel simulation to test the safety margin of different control strategies. An emergency broadcast unit is configured to send risk warnings to associated devices to trigger the collaborative protection mechanism of the surrounding systems.

[0029] Specifically, the edge-cloud data synchronization unit adopts differential compression technology, which reduces the communication bandwidth demand by more than 60% while ensuring data integrity by extracting the change amount between data frames instead of transmitting the full amount. The multi-level linkage arbitration unit builds a multi-party benefit equilibrium model based on game theory. When a thermal runaway risk is detected, the battery management system, charging pile controller and vehicle control unit jointly participate in decision-making, and the response priority and action timing of each system are determined through the Nash equilibrium algorithm. The digital twin mirror unit builds a high-fidelity battery virtual model in the cloud, which integrates electrochemical-thermal-mechanical coupling simulation engines, can test the safety boundaries of thousands of control strategies in parallel, and synchronize the optimization results to the physical system. The emergency broadcast unit adopts a hierarchical warning protocol. When the risk level exceeds the threshold, encrypted instructions are broadcast to associated devices through a low-latency communication network, triggering multi-level cascading protection mechanisms such as charging pile emergency power-off and vehicle system entering safe mode, forming a three-dimensional safety protection system across devices.

[0030] A real-time monitoring and early warning method for electric bicycle lithium battery data, referring to Figure 2 , comprising the following steps: S1: obtaining a multi-source heterogeneous data set of a lithium battery system, the multi-source heterogeneous data set including single cell voltage ripple spectrum, battery pack temperature field distribution matrix, charge and discharge current harmonic component and battery internal resistance dynamic change curve; S2: performing cross-domain feature alignment on the multi-source heterogeneous data set based on a space-time fusion algorithm to generate a comprehensive evaluation parameter set including health state score, thermal runaway risk index and remaining life prediction value; S3: constructing an adaptive early warning boundary model according to the comprehensive evaluation parameter set, and calculating voltage balance compensation coefficient, maximum allowable charging rate correction factor and thermal gradient suppression parameter through a double-layer neural network; S4: generating a hierarchical control instruction set based on a multi-objective optimization algorithm, the instruction set including pulse charging duty cycle adjustment amount, active balancing topology switching strategy, cooling fan speed curve and battery state of charge calibration parameter; S5: synchronizing the hierarchical control instruction set to edge computing nodes and cloud management platforms, and triggering a multi-level linkage protection mechanism based on game theory when the thermal runaway risk index exceeds the first dynamic threshold.

[0031] The above are preferred embodiments of the present application, and are not intended to limit the protection scope of the present application, therefore: any equivalent changes made in structure, shape, principle, etc. according to the present application should be covered within the protection scope of the present application.

Claims

1. A real-time monitoring and early warning system for lithium battery data of electric bicycles, characterized in that: Includes the following modules: A multi-dimensional data acquisition module is used to obtain a multi-source heterogeneous data set of the lithium battery system, which includes a single cell voltage ripple spectrum, a battery pack temperature field distribution matrix, charge and discharge current harmonic components, and a dynamic change curve of the battery internal resistance; A collaborative feature extraction module is used to perform cross-domain feature alignment on the multi-source heterogeneous data set based on a spatiotemporal fusion algorithm to generate a comprehensive evaluation parameter set including a health status score, a thermal runaway risk index, and a remaining life prediction value; A dynamic threshold generation module is used to construct an adaptive warning boundary model based on the comprehensive evaluation parameter set, and calculate the voltage equalization compensation coefficient, the maximum allowable charging rate correction factor and the thermal gradient suppression parameter through a two-layer neural network; An intelligent decision-making module for generating a hierarchical control instruction set based on a multi-objective optimization algorithm. The instruction set includes pulse charging duty cycle adjustment, active balancing topology switching strategy, cooling fan speed curve, and battery state of charge calibration parameters; The cloud collaboration module is used to synchronize the hierarchical control instruction set to the edge computing node and the cloud management platform, and trigger a multi-level linkage protection mechanism based on game theory when it is detected that the thermal runaway risk index exceeds the first dynamic threshold.

2. The electric bicycle lithium battery data real-time monitoring and early warning system according to claim 1 is characterized in that: The collaborative feature extraction module includes: a ripple characteristic analysis unit, configured to perform a wavelet transform-Hilbert joint analysis on the cell voltage ripple spectrum to extract the high-frequency oscillation energy accumulation value and the low-frequency drift component. The ripple characteristic analysis unit distinguishes normal charge and discharge ripple from abnormal oscillation caused by early faults by establishing a voltage fluctuation energy density spectrum. A primary warning is triggered when the high-frequency oscillation energy accumulation value exceeds a preset reference value. a thermal field modeling unit, configured to establish a three-dimensional heat conduction equation for the temperature field distribution matrix, solve the heat flux distribution within the battery pack using a finite element method, and calculate a critical heat flux density under a maximum temperature difference constraint. The thermal field modeling unit compares the deviation between the actual heat flux density and the critical value in real time, and initiates an active heat dissipation strategy when the critical heat flux density threshold in a local hotspot exceeds a preset threshold; An impedance spectrum analysis unit is used to fit the relaxation time constant of the dynamic change curve of the internal resistance by the least squares method and solve it using the Levenberg-Marquardt optimization algorithm. The impedance spectrum analysis unit simultaneously constructs a Cole-Cole spectrum including multiple characteristic frequency points, and dynamically evaluates the battery aging state by monitoring the change rate of the relaxation time constant. When the sudden change of the relaxation time constant exceeds two standard deviations of the historical mean, it is determined to be abnormal aging; The feature fusion subunit is used to perform weighted fusion of the above three types of features using the attention mechanism, and the weight coefficient is dynamically adjusted according to the current working conditions.

3. The electric bicycle lithium battery data real-time monitoring and early warning system according to claim 2 is characterized in that: The dynamic threshold generation module includes: A nonlinear mapping unit is used to construct a two-dimensional decision plane including the health state score SOH and the thermal runaway risk index R_th, and generate a dynamic warning boundary surface; the surface divides the working state into three areas: a safe zone, a warning zone, and a dangerous zone. When a data point enters the warning zone, a derated operation mode is activated; An adaptive learning unit is configured to construct a two-layer neural network using a long short-term memory (LSTM) network. The weight matrix W is updated online based on the battery degradation trajectory, and the update rule incorporates a dynamic ripple energy adjustment factor. The loss function includes voltage consistency error, temperature gradient error, and capacity decay error. The adaptive learning unit performs global parameter updates every preset number of charge and discharge cycles, while retaining a copy of the historical optimal weights as a rollback backup. The threshold optimization subunit is used to generate a group of random operating condition data through Monte Carlo simulation and optimize the combination of each warning threshold using particle swarm algorithm.

4. The electric bicycle lithium battery data real-time monitoring and early warning system according to claim 1 is characterized in that: The intelligent decision-making module includes: A multi-objective optimization unit is used to construct a dynamic optimization space in the dimensions of voltage balance, temperature stability, and battery life loss. It automatically adjusts the optimization weights based on the user-defined performance priority, safety priority, or life priority mode and outputs the optimal control parameter combination. A pulse charging control unit uses an adaptive fuzzy proportional-integral-derivative (PID) algorithm to adjust the charging current waveform in real time. It dynamically selects multiple preset pulse modes based on battery temperature and health status. Each pulse cycle includes an intelligently adjusted charging period and relaxation interval. The balancing strategy decision unit is used to intelligently switch between active and passive balancing modes by monitoring the battery pack voltage difference and temperature gradient in real time. Active balancing uses a distributed energy transfer architecture based on a switched capacitor matrix, and the balancing accuracy is controlled within a preset range. The thermal management control unit is used to generate a hierarchical cooling strategy based on the results of temperature field analysis, control the cooling fan speed through pulse width modulation (PWM), and activate directional strong cooling mode when local hot spots are detected.

5. The electric bicycle lithium battery data real-time monitoring and early warning system according to claim 1 is characterized in that: The cloud collaboration module includes: Edge-to-cloud data synchronization unit, which uses differential compression technology to achieve two-way real-time synchronization of control instructions and status data, ensuring instruction consistency between edge nodes and cloud platforms; A multi-level linkage arbitration unit is used to build a collaborative decision-making model among the battery pack, charging station, and vehicle system based on game theory when the risk of thermal runaway exceeds the limit, dynamically allocating the protection response priority of each system; A digital twin mirror unit is used to build a virtual model in the cloud that is fully synchronized with the physical battery, and is used for large-scale parallel simulation to test the safety margins of different control strategies; The emergency broadcast unit is used to send risk warnings to related devices and trigger the coordinated protection mechanism of surrounding systems.

6. A real-time monitoring and early warning method for lithium battery data of electric bicycles, characterized in that: The following steps are involved: Acquire a multi-source heterogeneous data set of a lithium battery system, wherein the multi-source heterogeneous data set includes a single cell voltage ripple spectrum, a battery pack temperature field distribution matrix, a charge and discharge current harmonic component, and a battery internal resistance dynamic change curve; Performing cross-domain feature alignment on the multi-source heterogeneous data set based on a spatiotemporal fusion algorithm to generate a comprehensive evaluation parameter set including a health status score, a thermal runaway risk index, and a remaining life prediction value; An adaptive warning boundary model is constructed based on the comprehensive evaluation parameter set, and a voltage equalization compensation coefficient, a maximum allowable charging rate correction factor, and a thermal gradient suppression parameter are calculated through a two-layer neural network; Generate a hierarchical control instruction set based on a multi-objective optimization algorithm, the instruction set including pulse charging duty cycle adjustment, active balancing topology switching strategy, cooling fan speed curve, and battery state of charge calibration parameters; The hierarchical control instruction set is synchronized to the edge computing node and the cloud management platform, and when it is detected that the thermal runaway risk index exceeds the first dynamic threshold, a multi-level linkage protection mechanism based on game theory is triggered.

7. The electric bicycle lithium battery data real-time monitoring and early warning method according to claim 6 is characterized in that: The steps of performing cross-domain feature alignment on the multi-source heterogeneous data set based on a spatiotemporal fusion algorithm to generate a comprehensive evaluation parameter set including a health status score, a thermal runaway risk index, and a remaining life prediction value are specifically as follows: A wavelet transform-Hilbert joint analysis is performed on the monomer voltage ripple spectrum to extract the high-frequency oscillation energy accumulation value and the low-frequency drift component. The ripple feature analysis unit distinguishes normal charge and discharge ripple from abnormal oscillation caused by early faults by establishing a voltage fluctuation energy density spectrum. When the high-frequency oscillation energy accumulation value exceeds a preset reference value, a primary warning is triggered. A three-dimensional heat conduction equation is established for the temperature field distribution matrix, and a finite element method is used to solve the heat flux distribution inside the battery pack. The critical heat flux density under the maximum temperature difference constraint is calculated. The thermal field modeling unit compares the deviation between the actual heat flux density and the critical value in real time, and activates an active heat dissipation strategy when the critical heat flux density threshold of the local hot spot exceeds a preset threshold. The relaxation time constant of the dynamic change curve of the internal resistance is fitted by the least squares method and solved by the Levenberg-Marquardt optimization algorithm. The impedance spectrum analysis unit simultaneously constructs a Cole-Cole spectrum including multiple characteristic frequency points. The dynamic evaluation of the battery aging state is achieved by monitoring the change rate of the relaxation time constant. When the sudden change of the relaxation time constant exceeds two standard deviations of the historical mean, it is determined to be abnormal aging; The attention mechanism is used to perform weighted fusion of the above three types of features, and the weight coefficient is dynamically adjusted according to the current working conditions.

8. The electric bicycle lithium battery data real-time monitoring and early warning method according to claim 7 is characterized in that: The steps of constructing an adaptive warning boundary model based on the comprehensive evaluation parameter set and calculating the voltage equalization compensation coefficient, the maximum allowable charging rate correction factor and the thermal gradient suppression parameter through a two-layer neural network are specifically as follows: A two-dimensional decision plane consisting of the health status score (SOH) and the thermal runaway risk index (R_th) is constructed to generate a dynamic warning boundary surface. This surface divides the operating state into three zones: a safe zone, a warning zone, and a dangerous zone. When a data point enters the warning zone, a derated operation mode is initiated. A two-layer neural network is constructed using a long short-term memory (LSTM) network. Its weight matrix W is updated online based on the battery degradation trajectory, and the update rule introduces a dynamic ripple energy adjustment factor. The loss function includes voltage consistency error, temperature gradient error, and capacity decay error. The adaptive learning unit performs global parameter updates every preset charge and discharge cycle, while retaining a copy of the historical optimal weight as a rollback backup. A group of random operating condition data is generated through Monte Carlo simulation, and the particle swarm algorithm is used to optimize the combination of each warning threshold.

9. The electric bicycle lithium battery data real-time monitoring and early warning method according to claim 6, characterized in that: A hierarchical control instruction set is generated based on a multi-objective optimization algorithm. The instruction set includes the steps of adjusting the pulse charging duty cycle, the active balancing topology switching strategy, the cooling fan speed curve, and the battery state of charge calibration parameters. Specifically, Build a dynamic optimization space based on voltage balance, temperature stability, and battery life loss, automatically adjust the optimization weights based on the user-defined performance priority, safety priority, or life priority mode, and output the optimal control parameter combination; Adopting an adaptive fuzzy proportional-integral-derivative (PID) algorithm to adjust the charging current waveform in real time, the system dynamically selects multiple preset pulse modes based on battery temperature and health status. Each pulse cycle includes intelligently adjusted charging period and relaxation interval. By real-time monitoring of battery pack voltage differences and temperature gradients, it intelligently switches between active and passive balancing modes. Active balancing uses a distributed energy transfer architecture based on a switched capacitor matrix, and the balancing accuracy is controlled within a preset range. A hierarchical cooling strategy is generated based on the temperature field analysis results. The cooling fan speed is controlled through pulse width modulation (PWM), and a directional strong cooling mode is activated when a local hot spot is detected.

10. The electric bicycle lithium battery data real-time monitoring and early warning method according to claim 6, characterized in that: Based on the temperature field analysis results, a hierarchical cooling strategy is generated. The cooling fan speed is controlled through pulse width modulation (PWM). When a local hotspot is detected, a directional strong cooling mode is activated. The specific steps are as follows: Differential compression technology is used to achieve two-way real-time synchronization of control instructions and status data, ensuring instruction consistency between edge nodes and cloud platforms; When the risk of thermal runaway exceeds the limit, a collaborative decision-making model among the battery pack, charging pile, and vehicle system is constructed based on game theory to dynamically allocate the protection response priority of each system. Build a virtual model in the cloud that is fully synchronized with the physical battery for large-scale parallel simulation testing of the safety margins of different control strategies; Send risk warnings to related devices and trigger the coordinated protection mechanism of surrounding systems.

Images